高超声速飞行器转捩高维数据高效辨识与智能CFD预报方法

Accurate prediction for the area and procedure of boundary transition is very important to the propulsion system, thermo protective performance, aerodynamic configuration and flight control of near-space hypersonic vehicles. However, current researches on hypersonic transition models can only apply to simple aerodynamic configure and limited flight conditions. Moreover, it is difficult for these models to be improved with data of flight tests. On the other hand, wind tunnel tests are very costly as well as inaccurate in simulating the hypersonic flight environment. Therefore, accurate simulation and prediction for hypersonic transition have become a worldwide challenging problem in both aerospace and mechanics field. Combining computational fluid dynamics (CFD) approaches with flight tests and intelligent modeling & identification approaches, this project studies a hybrid intelligent prediction approach for hypersonic transitions. A multi-partitioned parallel numerical simulation approach is proposed for hypersonic transition / turbulent flow field of complex aerodynamic configuration. Moreover, to be accurate in various flight states and boundary conditions, an intelligent CFD approach based on an intelligent reasoning system is studied for RANS (Reynolds Average Navier-Stockes) models selection and RANS coefficients estimation. With amass high dimension data of flow field from large eddy simulation and incomplete flight test data, an efficient intelligent approach will be developed to identify RANS coefficients based on a semi-supervised learning algorithm for decreasing data dimension and a modeling error probability density function (PDF) control algorithm. The research findings will be tested and validated in National Major Project on Near-space Hypersonic Vehicle. Finally, this project will deliver hybrid intelligent prediction models, algorithms and software for the development of hypersonic vehicles, which can help the improvement of the precision, practicality and efficiency of hypersonic boundary transition prediction as well as have great significance of theory and application.

准确预报高超声速边界层转捩区域和过程对飞行器推进系统、防隔热设计、气动布局、飞行控制等至关重要。现有转捩预报方法仅适用简单飞行器外形和有限状态，且难以结合飞行试验改进，风洞试验成本高且难以模拟真实飞行环境和主要参数影响，因而成为力学与航天领域世界性难题。本项目将计算流体力学(CFD)与飞行试验和智能建模、辨识方法相结合，针对高超声速飞行器复杂外形，研究基于多分区混合计算和并行数值模拟的模式法转捩仿真模型，并建立模式常数估计和预报模型选择的智能推理系统，实现适应不同飞行工况和边界条件的智能CFD预报方法；针对高维海量流场数据、信息不完整试验数据和转捩模型复杂计算过程，研究基于半监督学习降维和误差概率密度控制的模式常数高效智能辨识方法。集成本项目研究内容，研制高超声速转捩混合智能预报系统，应用、验证于高超声速飞行器重大科技工程，提高预报准确性、实用性和效率，具有重要科学意义和应用价值。

高超声速飞行器流场转捩过程可严重影响飞行安全和气动布局、增大热防护系统负担、降低飞行器有效载荷。现有的转捩高精度数值模拟方法、模式转捩方法和工程估算方法仅适用简单飞行器外形和极为有限的飞行状态，难以应用于航天工程指导飞行器设计及验证。并且，难以通过风洞试验来模拟转捩过程或评估其影响。因此，本项目针对这一力学与航天领域世界性难题，提出了一种智能建模和并行数值模拟技术相结合的高超声速飞行器转捩混合智能预报方法，主要研究了：1）基于模式法的转捩/湍流流场多分区并行数值模拟方法；2）基于流场高维数据的转捩模式常数高效辨识方法；3）基于转捩模式常数估计的智能CFD预报方法；4）基于高超声速飞行器重大科技工程的转捩预报应用。通过本项目的实施，建立了面向高超声速飞行器复杂外形的多分区混合模拟转捩仿真模型和高维数据驱动智能CFD模型，能够预测转捩发生的位置，确定转捩区的长度，已完成了课题立项时计划的研究目标和预期研究成果。主要研究成果，在国内外已公开发表论文20篇、获批软件版权2项、申请专利5项（正在审核），在研期间培养3名博士研究生，3名参与人晋升副高级职称，研制的算法形成软件已集成入中国运载火箭技术研究院的Hyper-CFD气动流场模拟平台，支撑型号工程和预先研究的高超声速飞行器设计。本项目研究的转捩分区并行模拟方法、高维数据驱动智能CFD预报方法，已应用于国家临近空间高超声速飞行器重大科技工程，结合试验数据、数值模拟数据等研制预报模型，准确高效地预报转捩过程，建立了将流场数值模拟与数据驱动智能建模相融合解决气动设计/验证问题的研究模式，提升了设计水平和飞行安全性。同时，本项目融合数据驱动智能建模、机器学习与气动流场数值模拟或工程算法的混合智能建模/预报方法，已推广应用于高超声速装备智能化、强机动化的改进工程。